Deep learning based underwater sonar image target recognition has been a popular research direction. However, target recognition in sonar images continues to pose significant challenges when compared to target recognition research involving optical images. To address this issue, this study aims to propose an underwater sonar image target recognition model based on YOLOv8, which is specifically crafted to enhance the accuracy of target recognition in complex underwater scenarios. This paper addresses the issue of low resolution in sonar images by introducing the SPD-Conv CNN module, designed specifically for low-resolution images. The integration of this module leads to a significant improvement of mean average precision (mAP) by 1% when compared to the original model. Further, the Coordinate attention module leads to an additional 1.2% improvement in the recognition performance compared to the original model. Additionally, we replace the original activation function with Gaussian Error Linear Unit, resulting in a further performance enhancement of 1.9% when compared to the original model. Overall, all these model improvements collectively improved the mAP by 3.1% to reach 98.4%. The experimental results show that our model has excellent performance in target recognition in underwater sonar images.